International Journal of Engineering Research and Developmente-ISSN: 2278-067X, p-ISSN : 2278-800X, www.ijerd.comVolume 4,...
Pattern Recognition System using MLP Neural NetworksMLP learn through an iterative process of adjustments applied to their...
Pattern Recognition System using MLP Neural NetworksThe activation functions used were hyperbolic tangent for the hidden l...
Pattern Recognition System using MLP Neural NetworksTable - 2 for Input Statistical         Parameters             46
Upcoming SlideShare
Loading in …5
×

Welcome to International Journal of Engineering Research and Development (IJERD)

218 views

Published on

http://www.ijerd.com

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
218
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
3
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Welcome to International Journal of Engineering Research and Development (IJERD)

  1. 1. International Journal of Engineering Research and Developmente-ISSN: 2278-067X, p-ISSN : 2278-800X, www.ijerd.comVolume 4, Issue 9 (November 2012), PP. 43-46 Pattern Recognition System using MLP Neural Networks Sarvda Chauhan1, Shalini Dhingra2 1,2 ECE Deptt., IET, Bhaddal, Ropar, Punjab INDIA, Abstract :- Many machine vision researchers are working on the area of pattern recognition, a wide variety of approaches have been proposed to attempt to achieve the recognition system. These approaches generally fall into two categories: statistical method and syntactic method. MLP is an ideal means of tackling whole range of difficult tasks in pattern recognition and regression because of its highly adaptable non-linear structure. The trend of using multi-layer perceptron neural network for the solution of pattern recognition problem is understandable due to their capacity to accommodate the nature of human brain learning capacity, and the fact that their structure can be formulated mathematically. The learning process takes place through adjustment of weight connections between the input and hidden layers and between the hidden and output layers. The number of output nodes in the proposed work are circle, half circle, equilateral triangle, square and hexagon and the number of nodes in the hidden layer are chosen based on trial and error method. The different parameters like mean radius, Standard Deviation of Radii, Normalized Area, Perimeter, Mean of Max. Radii & Mean of Min Radii are considered to check the effectiveness of the proposed algorithm. I. INTRODUCTION Layered neural networks involve a set of input cells connected to a collection of output cells by means of one ormore layers of modifiable intermediate connections [1]. The most natural use of this architecture is to implementassociatively by associating an input pattern, the representation of a concept or a situation, with another item, either of thesame kind or totally different. For this reason, networks with multiple layers are described as associative networks. Patternrecognition is the primary use of this mechanism. Pattern recognition is a popular application in that it enables the full set ofhuman perceptions to be acquired by a machine. In the context of image processing, the different stages are: acquisition,concern with digitizing the data coming from a sensor such as camera or scanner, localizing, which involves extracting theobject from its background, and representation. Some features are extracted from the input pattern and that becomes theinput to neural network. The final stage, the decision stage, consists of dividing the set of object representations into numberof classes. The last two phases, representation and decision, are the associative phases. Layered neural networks can play apart in each of these phases. II. BRIEF LITERATURE SURVEY Multilayer Feed-forward Neural Networks (MFNNs) is an ideal means of tackling a whole range of difficult tasksin pattern recognition and regression because of its highly adaptable non-linear structure. The trend of using multi-layerperceptron neural network [1] for the solution of pattern recognition problem is understandable due to their capacity toaccommodate the nature of human brain learning capacity, and the fact that their structure can be formulated mathematically.The functionality of the topology of the MLP is determined by a learning algorithm able to modify the parameters of the net.The algorithm of back propagation, based on the method of steepest descent [4] in the process of upgrading the connectionweights, is the most commonly used by the scientific community. Multi-Layer Perceptrons (MLP) are fully connected feedforward nets with one or more layers of nodes between the input and the output nodes. Each layer is composed of one ormore artificial neurons in parallel. III. FEATURE EXTRACTION Feature extraction is a process of studying and deriving useful information from the input patterns. In imagerecognition, the extracted features contain information about statistical parameters and are computed about the centre ofgravity of input pattern. The statistical parameters used in neural network training process are: Perimeter, Area, Maximumand Minimum radii in each quadrant, Intercepts on axes, Mean Radius, Standard Deviation and Figure Aspect. All theparameters discussed here are normalized with respect to the mean radius. The normalization step make the parameters sizeindependent. The statistical parameters so computed above are also made rotation independent using the orthogonaltransformation of the coordinates. IV. METHODOLOGY An input pattern is presented to the first stage of the forward paths, the input layer K, which consists of a two-dimensional array of receptor cells. The second stage of forward paths is called the hidden layer J. The third stage of theforward paths is the recognition layer I as shown in below figure. After the process of learning ends, the final result of thepattern recognition shows- in the response of the cells of FA. Let x1,x2,…,xn be the input signals and w1, w2, w3, ,…, wnbe the synaptic weights, and u the activation potential, & the threshold and y the output signal and f the activation function: 43
  2. 2. Pattern Recognition System using MLP Neural NetworksMLP learn through an iterative process of adjustments applied to their free parameters. The most common learningalgorithms are the standard back-propagation and faster-learning variations . They use a gradient search technique tominimize a cost function equal to the mean square error (MSE) between the desired and the actual net outputs.The net is trained by initially selecting small random weights and internal thresholds, and presenting all training datarepeatedly. Weights are adjusted after every trial usingInformation specifying the correct class until weights converges and the cost function is reduced to an acceptable value. Typical Architecture of a MLP Neural Network System V. EXPERIMENT The learning process takes place through adjustment of weight connections between the input and hidden layersand between the hidden and output layers. The number of input nodes at the input layer was set at 17 according to thestatistical features discussed in section II. Followings are the input nodes:1. Mean Radius2. Standard Deviation of Radii3. Normalized Area4. Perimeter5. Mean of Max. Radii6. Mean of Min RadiiThe number of output nodes in this study was set at 6 as circle, half circle, equilateral triangle, square and hexagon. Thenumber of nodes in the hidden layer was chosen based on trial and error. Each pattern presentation is tagged with itsrespective label as shown in below table. The maximum value in each row (0.9) identifies the corresponding node expectedto secure the highest output for a pattern to be considered correctly classified. The output values are denoted as O1, O2, O3,O4, O5 and O6. Target Des. Nodes Class 1 2 3 4 5 6 1 Square 0.9 0.1 0.1 0.1 0.1 0.1 2 Circle 0.1 0.9 0.1 0.1 0.1 0.1 3 Half 0.1 0.1 0.9 0.1 0.1 0.1 Circle 4 Eq. 0.1 0.1 0.1 0.9 0.1 0.1 Triangle 5 Reg. 0.1 0.1 0.1 0.1 0.9 0.1 Hexagon 6 Reg. 0.1 0.1 0.1 0.1 0.1 0.9 Pentagon 44
  3. 3. Pattern Recognition System using MLP Neural NetworksThe activation functions used were hyperbolic tangent for the hidden layer and sigmoid for the output layer. The hyperbolictangent function is given by:The sigmoid function is given by:The input node parameters are normalized to the level [0, 1]. 30 no. of patterns (5 from each category) data sets are used totrain the NN with a target of MSE < 0.01 as acceptable criteria for stopping of the training process. VI. RESULTS The NN is trained using the ANN tool box in MATLAB. The set of statistical parameters used are extracted usingthe feature extraction process. Table-2 gives the set of parameters for the respective shapes. Following results are obtainedafter the training is complete and new test pattern data sets are provided for testing of the classifier: Percentage Correct Target Class MSE Classification 1 - Square 96.78 0.0099 2 - Circle 97.45 0.0098 3 – Half Circle 98.34 0.0101 4 – Eq. Triangle 96.64 0.0090 5 – Reg. Hexagon 97.34 0.0092 6 – Reg. Pentagon 98.56 0.0097 VII. CONCLUSION Pattern recognition can be done both in normal computers and neural networks. Computers use conventionalarithmetic algorithms to detect whether the given pattern matches an existing one. It is a straight forward method. It will sayeither yes or no. It does not tolerate noisy patterns. On the other hand, neural networks can tolerate noise and, if trainedproperly, will respond correctly for unknown patterns. Neural networks may not perform miracles, but if constructed withthe proper architecture and trained correctly with good data, they will give amazing results, not only in pattern recognitionbut also in other scientific and commercial applications. REFERENCES[1]. A Versatile Machine Vision System for Complex Industrial Parts (IEEE Trans. On Computers, Sep. 1977, No. 9, Vol. C-27)[2]. Wani, M. A. and Pham, D. T., 1999, Efficient control chart pattern recognition through synergistic and distributed artificial neural networks. Proceedings of the Institution of Mechanical Engineers: Part B, 213, 157–169[3]. Anil k. Jain Fellow IEEE, Robert P.W Duin & Jian chang Mao, Senior Member IEEE “IEEE Transactions on Pattern Analysis and Machine Intelligence VOL 22 No.1, January 2000[4]. Enrique Frias –Marrtinez, Angel Sanchez & Jose Velez “Support Vector Machines Vs Multi-Layer Perceptrons for Efficient Off-Line Signature, Bio Metry & Artificial Vision Group[5]. Aki Vehtari and Juko Lampinen Bayesian MLP Neural Networks for Image Analysis” Laborator Of Computational engineering ,Helsinki University Of Technology, Finland 2000[6]. Yas Abbas Alsultanny, Musbah M. Aqel , 2002,Pattern recognition using multilayer neural-genetic Algorithm.. Elsevier Science B.V. Nuerocomputing 51(2003) 237-247[7]. Richard Bridge, “Computer Image Processing.” Tessella support services PLC, Oxon, England, Issue V1.R2.M0, 2003[8]. Raul C. Mauresan “Pattern Recognition Using Pulse coupled Neural Neural Network & Discrete Fourier Transforms “Personal Research[9]. NI Vision Based Automatic Optical Inspection (AOI), Proceedings of 2009 IEEE International Conference on Applied Superconductivity and Electromagnetic Devices Chengdu, China, September 25-27, 2009 45
  4. 4. Pattern Recognition System using MLP Neural NetworksTable - 2 for Input Statistical Parameters 46

×